DIBR 中基于平面扫描的深度重建方法

摘要

A novel plane-sweep based depth reconstruction method for depth image-based rendering is pres-ented. Unlike depth from stereo and image-based visual hull techniques, our method further develops the plane-sweeping algorithm, and extends it to synthesize "virtual" views of a real-world scene associated with depth information without any prior geometric information. When projecting the input images into the parallel "virtual depth planes", a dynamic scoring method to compute the color consistency is introduced.In addition, for building a photorealistic "virtual" view, we present a view-direction based weighting strategy for depth image-based rendering. An important advantage of our approach is that due to more ac- curate depth information, better novel views can be acquired.%提出一种 DIBR 中基于平面扫描法的深度重建方法,与立体深度重建算法和基于图像的视觉壳算法不同,本文进一步改进平面扫描算法,无需任何场景的几何先验知识,而是利用每个像素点的深度信息合成真实场景的虚拟视点.当输入图像映射至相互平行的虚拟深度平面时,采用"动态判决方法"来计算像素间的色彩一致度;并在虚拟视合成中采用了基于视向权重策略的新视点重建方法.本文算法获取的深度信息更为精确,虚拟新视点的质量得到较大提高.